Add the rest of AVX SSE4.1 packed move with sign/zero extend instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@107723 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 112 insertions, 0 deletions

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 7c678a67d4..264ec825bb 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -4370,6 +4370,17 @@ multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
           OpSize;
 }
 
+let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in {
+defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
+                                     VEX;
+defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
+                                     VEX;
+defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
+                                     VEX;
+defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
+                                     VEX;
+}
+
 defm PMOVSXBD   : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
 defm PMOVSXWQ   : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
 defm PMOVZXBD   : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
@@ -4400,6 +4411,12 @@ multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
                  OpSize;
 }
 
+let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in {
+defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
+                                     VEX;
+defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
+                                     VEX;
+}
 defm PMOVSXBQ   : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
 defm PMOVZXBQ   : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
 
diff --git a/test/MC/AsmParser/X86/x86_32-encoding.s b/test/MC/AsmParser/X86/x86_32-encoding.s
index 34792a901f..af905023d0 100644
--- a/test/MC/AsmParser/X86/x86_32-encoding.s
+++ b/test/MC/AsmParser/X86/x86_32-encoding.s
@@ -11998,3 +11998,51 @@
 // CHECK: encoding: [0xc4,0xe2,0x79,0x35,0x10]
           vpmovzxdq  (%eax), %xmm2
 
+// CHECK: vpmovsxbq  %xmm2, %xmm5
+// CHECK: encoding: [0xc4,0xe2,0x79,0x22,0xea]
+          vpmovsxbq  %xmm2, %xmm5
+
+// CHECK: vpmovsxbq  (%eax), %xmm2
+// CHECK: encoding: [0xc4,0xe2,0x79,0x22,0x10]
+          vpmovsxbq  (%eax), %xmm2
+
+// CHECK: vpmovzxbq  %xmm2, %xmm5
+// CHECK: encoding: [0xc4,0xe2,0x79,0x32,0xea]
+          vpmovzxbq  %xmm2, %xmm5
+
+// CHECK: vpmovzxbq  (%eax), %xmm2
+// CHECK: encoding: [0xc4,0xe2,0x79,0x32,0x10]
+          vpmovzxbq  (%eax), %xmm2
+
+// CHECK: vpmovsxbd  %xmm2, %xmm5
+// CHECK: encoding: [0xc4,0xe2,0x79,0x21,0xea]
+          vpmovsxbd  %xmm2, %xmm5
+
+// CHECK: vpmovsxbd  (%eax), %xmm2
+// CHECK: encoding: [0xc4,0xe2,0x79,0x21,0x10]
+          vpmovsxbd  (%eax), %xmm2
+
+// CHECK: vpmovsxwq  %xmm2, %xmm5
+// CHECK: encoding: [0xc4,0xe2,0x79,0x24,0xea]
+          vpmovsxwq  %xmm2, %xmm5
+
+// CHECK: vpmovsxwq  (%eax), %xmm2
+// CHECK: encoding: [0xc4,0xe2,0x79,0x24,0x10]
+          vpmovsxwq  (%eax), %xmm2
+
+// CHECK: vpmovzxbd  %xmm2, %xmm5
+// CHECK: encoding: [0xc4,0xe2,0x79,0x31,0xea]
+          vpmovzxbd  %xmm2, %xmm5
+
+// CHECK: vpmovzxbd  (%eax), %xmm2
+// CHECK: encoding: [0xc4,0xe2,0x79,0x31,0x10]
+          vpmovzxbd  (%eax), %xmm2
+
+// CHECK: vpmovzxwq  %xmm2, %xmm5
+// CHECK: encoding: [0xc4,0xe2,0x79,0x34,0xea]
+          vpmovzxwq  %xmm2, %xmm5
+
+// CHECK: vpmovzxwq  (%eax), %xmm2
+// CHECK: encoding: [0xc4,0xe2,0x79,0x34,0x10]
+          vpmovzxwq  (%eax), %xmm2
+
diff --git a/test/MC/AsmParser/X86/x86_64-encoding.s b/test/MC/AsmParser/X86/x86_64-encoding.s
index ef274cf0a4..62aebdba6a 100644
--- a/test/MC/AsmParser/X86/x86_64-encoding.s
+++ b/test/MC/AsmParser/X86/x86_64-encoding.s
@@ -2046,4 +2046,51 @@ pshufb	CPI1_0(%rip), %xmm1
 // CHECK: encoding: [0xc4,0x62,0x79,0x35,0x20]
           vpmovzxdq  (%rax), %xmm12
 
+// CHECK: vpmovsxbq  %xmm12, %xmm10
+// CHECK: encoding: [0xc4,0x42,0x79,0x22,0xd4]
+          vpmovsxbq  %xmm12, %xmm10
+
+// CHECK: vpmovsxbq  (%rax), %xmm12
+// CHECK: encoding: [0xc4,0x62,0x79,0x22,0x20]
+          vpmovsxbq  (%rax), %xmm12
+
+// CHECK: vpmovzxbq  %xmm12, %xmm10
+// CHECK: encoding: [0xc4,0x42,0x79,0x32,0xd4]
+          vpmovzxbq  %xmm12, %xmm10
+
+// CHECK: vpmovzxbq  (%rax), %xmm12
+// CHECK: encoding: [0xc4,0x62,0x79,0x32,0x20]
+          vpmovzxbq  (%rax), %xmm12
+
+// CHECK: vpmovsxbd  %xmm12, %xmm10
+// CHECK: encoding: [0xc4,0x42,0x79,0x21,0xd4]
+          vpmovsxbd  %xmm12, %xmm10
+
+// CHECK: vpmovsxbd  (%rax), %xmm12
+// CHECK: encoding: [0xc4,0x62,0x79,0x21,0x20]
+          vpmovsxbd  (%rax), %xmm12
+
+// CHECK: vpmovsxwq  %xmm12, %xmm10
+// CHECK: encoding: [0xc4,0x42,0x79,0x24,0xd4]
+          vpmovsxwq  %xmm12, %xmm10
+
+// CHECK: vpmovsxwq  (%rax), %xmm12
+// CHECK: encoding: [0xc4,0x62,0x79,0x24,0x20]
+          vpmovsxwq  (%rax), %xmm12
+
+// CHECK: vpmovzxbd  %xmm12, %xmm10
+// CHECK: encoding: [0xc4,0x42,0x79,0x31,0xd4]
+          vpmovzxbd  %xmm12, %xmm10
+
+// CHECK: vpmovzxbd  (%rax), %xmm12
+// CHECK: encoding: [0xc4,0x62,0x79,0x31,0x20]
+          vpmovzxbd  (%rax), %xmm12
+
+// CHECK: vpmovzxwq  %xmm12, %xmm10
+// CHECK: encoding: [0xc4,0x42,0x79,0x34,0xd4]
+          vpmovzxwq  %xmm12, %xmm10
+
+// CHECK: vpmovzxwq  (%rax), %xmm12
+// CHECK: encoding: [0xc4,0x62,0x79,0x34,0x20]
+          vpmovzxwq  (%rax), %xmm12